35
Thus, without additional data, dependent censoring is not identifiable (Tsiatis,
1975). For instance, it is reasonable to believe that a company being acquired implies
that it has limited ability to survive on its own. But the degree of such dependency
is unclear.
One feasible way that researchers normally take under these circumstances is to
conduct a sensitivity analysis, i.e., to evaluate model outputs under various scenarios
formed by reasonable contemplations. In this research, we also take such approach to
assess the impact of dependent censoring on the performance of the proposed model
and to help answer questions such as,
1. Would some covariates become non-significant when dependent censoring is
taken into account?
2. Would the parameter estimates be affected when the correlation between the
dependent censoring and events changes?
Although there has been a wide range of sensitivity analysis methods being stud-
ied, we found many of them are difficult to apply to our research. It was Zheng
and Klein’s (1995) method that inspired us to utilize the copula approach to model
the structure of dependent censoring. We extended their method to the sensitivity
analysis under a general setting of the well-known Cox proportional hazards model.